amiralifirouzi/Cassandane-qwen3-4b-persian-lora - Каталог нейросетей
Генерация текста

amiralifirouzi/Cassandane-qwen3-4b-persian-lora

Добавлено:
amiralifirouzi/Cassandane-qwen3-4b-persian-lora

Cassandane (کاساندان) is a Persian-focused LoRA fine-tune of Qwen3-4B, designed for Persian conversation, instruction following, and text generation. The name refers to Cassandane (کاساندان), the Achaemenid queen and wife of Cyrus the Great. > This repository contains the LoRA adapter, not the full base model. Evaluation was performed zero-shot with greedy decoding. PerCul random baseline: 25%. The largest improvement is on PerCoR, where performance approximately doubled. 14,934 samples after quality filtering, consisting of Persian conversational and agentic/tool-use data. 764 examples contain explicit reasoning traces; therefore, Cassandane should not be considered a dedicated reasoning model. — Trained on ~15K examples; not intended as an encyclopedic Persian knowledge model. — Reasoning training is limited due to the small number of reasoning examples. — PerCul performance remains close to the random baseline. — Outputs may contain factual errors or hallucinations. — Do not rely on the model for medical, legal, financial, or other high-stakes decisions. — Base model: unsloth/Qwen3-4B — Training dataset: amiralifirouzi/tiny-persian-sft-15k — Model:…

Модальности:
Генерация текста

Области применения:
Диалог / чат


Задача: Генерация текста
Автор: amiralifirouzi
Теги: peft, persian, farsi, qwen3, lora, sft, conversational, fa
Лайков: 3  |  Загрузок: 51

Открыть на HuggingFace →

Описание основано на материалах HuggingFace. Перевод выполнен автоматически.